Medizinische Universität Graz - Research portal
Selected Publication:
Wagner, F.
Structural Brain Networks as Predictors of Cognitive Decline in Alzheimer’s Disease – A Longitudinal Study
Doktoratsstudium der Medizinischen Wissenschaft; Humanmedizin; [ Dissertation ] Medizinische Universitaet Graz; 2024. pp. 86
[OPEN ACCESS]
FullText
- Authors Med Uni Graz:
- Advisor:
- Fazekas Christian
- Koini Marisa
- Schmidt Reinhold
- Altmetrics:
- Abstract:
- Studying grey matter structural covariance networks (SCNs) may enhance our knowledge about Alzheimer’s disease (AD) beyond the information based on the assessment of isolated grey matter areas. We, therefore, aim to (1) identify networks with diagnostic power for the differentiation between AD patients and healthy controls (HC), (2) compare their diagnostic power to established markers (brain volume, hippocampus volume, medial temporal lobe atrophy-score) and (3) determine the networks association to cognitive ability and cognitive decline. To identify networks with diagnostic power, we trained a random forest model on a sample of 104 AD patients and 104 age- and sex-matched HC and calculated the variable importances for a set of 20 SCNs. To prevent overfitting, we validated the model on an independent sample of 28 AD patients from another centre and 28 HC. Resulting networks were compared against above mentioned established markers by combining them in an additional random forest model, again, by training and validating in independent cohorts. To determine the networks association to cognitive ability and cognitive decline, multiple linear regression models were used. Two of the 20 SCNs showed significant contribution to the discrimination between AD and HC. These two networks comprise a temporal SCN and a secondary somatosensory SCN showing diagnostic accuracy of 81% in the training set and 86% in the validation set. When compared with the already established markers, the SCN performed worse and did not add any further information in a combined model. We found the temporal SCN to be associated with verbal memory at baseline, but this effect vanished after controlling for age, sex and education. We conclude that SCNs have diagnostic potential, but they do not provide information beyond established clinical tools.